Portuguese Social Media Hate Speech Corpus with 43,668 Annotations

43,668 annotated Portuguese documents from social networks are compiled for hate speech detection. The dataset, created by Silly-Machine, is an enhanced version that integrates annotations from several prior academic works and includes 10,000 original entries. It was last updated on the platform in January 2024.

Use Cases

• Train a text classification model to predict hate speech labels from the annotated document text.
• Fine-tune a Portuguese BERT model on the annotated text data for improved contextual understanding of abusive language.
• Analyze linguistic patterns and features within the annotated text to identify common markers of hate speech in Portuguese social media.
• Benchmark the performance of different machine learning algorithms on the task of hate speech detection using the provided annotations.
• Study dataset integration and annotation consistency by comparing the labels from the merged source datasets (Fortuna et al., Leite et al., Vargas et al.).

Strengths

• Contains 43,668 annotated documents, providing a substantial corpus for model training.
• Integrates and expands upon multiple established academic datasets, potentially improving coverage and reliability.

Limitations

• Specific column names, data schema, and annotation guidelines are not detailed in the provided information.
• The dataset's geographic and demographic representativeness within Portuguese-speaking social media is unclear.

Provenance

Source

Aggregated from datasets by Fortuna et al. (2019), Leite et al. (2020), and Vargas et al. (2022), plus 10,000 original entries.

Collection Method

Crowdsourced annotations, compiled and enhanced by the author.

Freshness

Last updated on the Hugging Face platform on 2024-01-01.

Geography

Region tag indicates 'us', but content language is Portuguese; specific geographic coverage is unclear.